Axitinib effectively inhibits BCR-ABL1(T315I) with a distinct binding conformation

The BCR-ABL1 fusion gene is a driver oncogene in chronic myeloid leukaemia and 30–50% of cases of adult acute lymphoblastic leuk- aemia1. Introduction of ABL1 kinase inhibitors (for example, ima- tinib) has markedly improved patient survival2, but acquired drug resistance remains a challenge3–5. Point mutations in the ABL1 kinase domain weaken inhibitor binding6 and represent the most common clinical resistance mechanism. The BCR–ABL1 kinase domain gate- keepermutation Thr315Ile (T315I) confers resistance to all approved ABL1 inhibitors except ponatinib7,8, which has toxicity limitations. Here we combine comprehensive drug sensitivity and resistance pro- filing of patient cells ex vivo with structural analysis to establish the VEGFR tyrosine kinase inhibitor axitinib as a selective and effective inhibitor for T315I-mutant BCR–ABL1-driven leukaemia. Axitinib potently inhibited BCR–ABL1(T315I), at both biochemical andcellular levels, by binding to the active form of ABL1(T315I) in a mutation-selective binding mode. These findings suggest that the T315I muta- tion shifts the conformational equilibrium of the kinase in favour of an active (DFG-in) A-loop conformation, which has more optimal binding interactions with axitinib. Treatment of a T315I chronic myeloid leukaemia patientwithaxitinibresultedina rapid reduction of T315I-positive cells from bone marrow. Taken together, our find- ings demonstrate an unexpected opportunity to repurpose axitinib, an anti-angiogenic drug approved for renal cancer, as an inhibitor for ABL1 gatekeeper mutantdrug-resistantleukaemiapatients. This study shows that wild-type proteins do not always sample the con- formations available to disease-relevant mutant proteins and that comprehensive drug testing of patient-derived cells can identify unpre- dictable, clinically significant drug-repositioning opportunities.

Unexpectedly, a strong selective response was detected with the vas- cular endothelial growth factor (VEGF) receptor tyrosine kinase inhi- bitor axitinib. Highly selective for VEGFR14, axitinib is approved as an anti-angiogenic agent for treating renal cell carcinoma15. No other VEGFR inhibitors in our drug collection showed activity in these cells, suggest- ing that the effect of axitinib was due to inhibition of a different target. Since axitinib had not been previously investigated in cell-based sys- tems for potency and selectivity towards ABL1, we explored whether the observed ex vivo effects were due to direct targeting of the BCR– ABL1(T315I) kinase and characterized the biochemical interactions. Surprisingly, axitinib more potently inhibited the kinase activity of ABL1(T315I) (inhibition constant (Ki) 5 100 pM), in line with VEGFR2 potency (Ki 5 20 pM)14,16, in comparison to wild-type ABL1 (Ki 5 3,800 pM).

The only clinically available inhibitor that has shown efficacy against BCR–ABL1(T315I)-driven disease is the broad-spectrum kinase inhib- itorponatinib. Ina recentphase III clinical trial with ponatinib, frequent severely adverse vascular effects were observed leading to termination of the trial as well as temporary withdrawal from the market9. More- over, compound mutations in BCR–ABL1(T315I) have been reported to cause resistance to ponatinib10,11. Therefore, there is a significant, un- met need for safe and effective therapies for BCR–ABL1(T315I)-driven leukaemia.

We performed phenotypic drug sensitivity and resistance testing (DSRT)12,13 of primary cells from chronic myeloid leukaemia (CML) and Philadelphia-chromosome-positive (Ph1) acute lymphoblastic leukaemia (ALL) patients using a collection of 252 approved and inves- tigational oncology compounds (Supplementary Table 1). DSRT data froma Ph1 B-cell ALL (B-ALL) patient (FHRB.1278) carrying the T315I mutation showed that the patient cells were insensitive to imatinib, dasatinib and nilotinib, and sensitive to ponatinib (Fig. 1a, b and Sup- plementary Table 2). Anticipated cancer-selective responses were observed to drugs targeting key BCR–ABL1 downstream effector sig- nals such as PI(3)K/MTOR, MEK and BCL2 (Supplementary Table 2).

Crystal structures of axitinib bound to wild-type and T315I ABL1 were determined (Extended Data Table 1) to understand binding interac- tions and selectivity. The structures revealed a striking difference in the activation loop (A-loop) conformations. While the wild-type ABL1 bound axitinib in an inactive (DFG-out conformation), as has also been shown for the axitinib–VEGFR complex, the axitinib–ABL1(T315I) complex had an active (DFG-in) A-loop conformation (Fig. 2a, b).

As the T315I gatekeeper mutation is known to stabilize the active kinase conformation17,18, the difference in A-loop conformations be- tween the wild type and the T315I mutant probably reflects the altered protein dynamics. The higher potency against the mutant implies that axitinib binds better to the active ABL kinase conformation. This was corroborated with biochemical studies where axitinib more potently in- hibited autophosphorylated (Ki 5 149 pM) rather than non-phosphory- lated ABL1(T315I) (Ki 5 421 pM).

Moreover, axitinib filled different binding space than other ABL1 in- hibitors (ponatinib, imatinib, dasatinib, nilotinib, bosutinib), as it did not extend as far towards the gatekeeper residue and a-helix C (Fig. 2c). Also, unlike other ABL1 inhibitors that rely on forming a hydrogen bond with the T315 residue and thus lose potency with an isoleucine sub- stitution19, axitinib was not positioned to form a hydrogen bond with T315 and therefore did not clash unfavourably with I315. These find- ings reveal the structural underpinnings of potent, ABL1(T315I)- selective inhibition, which can be used in the rational design of a new generation of ABL1-directed drugs.

Unexpectedly, axitinib adopts a significantly different binding con- formation in the active site of ABL1 relative to VEGFR214, specifically to the P-loop and A-loop conformation of the kinases (Fig. 2d and Ex- tended Data Fig. 1). A large rotational difference of the axitinib sulfur– indazole bond placed the phenyl amide group indifferent binding pockets. Next, to test whether axitinib specifically blocks ABL1(T315I) kinase activity in cells we used murine pro-B Ba/F3 cells stably expressing ei- ther wild-type or mutant BCR–ABL1 to examine ABL1 autophosphor- ylation and BCR–ABL1-dependent cell proliferation. Similar to ponatinib, axitinib potently reduced autophosphorylation of ABL1(T315I) and potently blocked proliferation of Ba/F3 cells expressing BCR–ABL1(T315I) in a dose-dependent manner (Fig. 3a, b). Strikingly, axitinib inhibited T315I-mediated autophosphorylation and Ba/F3 growth with ,10-fold higher potency compared to wild-type ABL1 (Fig. 3c, d). Finally, in a pro- file against a large set of clinically relevant BCR–ABL1 drug-resistance mutations expressed in Ba/F3 cells, axitinib showed high selective in- hibitory activity towards gatekeeper mutations (Extended Data Fig. 2). Taken together, our data demonstrate that axitinib selectively and ef- fectively targets gatekeeper-mutant BCR–ABL1 on a biochemical, struc- tural and cellular level.

To assess the efficacy of axitinib in patient samples further, we per- formed DSRT on mononuclear cells isolated from the bone marrow of two CML patients negative for the T315I mutation as well as three addi- tional patients carrying the T315I mutation (patient characteristics de- scribed in Extended Data Table 2). When compared with a panel of 32 primary acute myeloid leukaemia samples and seven healthy bone mar- row controls, samples derived from patients with BCR–ABL1(T315I)- driven leukaemia showed a selective response to axitinib, confirming our observations with the Ph1 B-ALL patient (FHRB.1278) (Fig. 4a and Extended Data Table 3). In accordance with the biochemical and Ba/F3 cell viability results, axitinib selectively and effectively inhibited the vi- ability of three of four CML and Ph1 ALL primary cell samples har- bouring the T315I mutation with median IC50 values 20-fold lower (Fig. 4b and Supplementary Table 3). The axitinib-insensitive patient sample was also insensitive to ponatinib, suggesting that the patient had a BCR–ABL1-independent disease (Fig. 4c).

We then focused on one of the tested CML patients (FHRB.1408) for whom all approved treatment options had been exhausted and ex vivo cell viability analysis had shown a solid response to axitinib (Extended Data Fig. 3a, b). To determine whether axitinib could block BCR– ABL1(T315I)-mediated signalling in the patient’s tumour cells, we assessed phosphorylation of the BCR–ABL1 substrate CRKL. Ex vivo axitinib treatment caused a dose-dependent reduction of CRKL phos- phorylation, indicating that axitinib effectively inhibited BCR–ABL1 (T315I) signalling in the primary leukaemic cells (Fig. 4d). Based on the potent on-target ex vivo responses, the patient was treated with an approved therapeutic dose of axitinib (5 mg twice daily) for two weeks.

The treatment resulted in rapid clearance of BCR–ABL1(T315I)-pos- itive cells as determined by a reduction of the T315I transcript levels in bone marrow (Fig. 4e). This finding suggests that axitinib can produce specific and effective responses in patients with BCR–ABL1(T315I)- driven disease.
Although the majority of patients with chronic phase CML achieve a significant therapeutic benefit with clinically available BCR–ABL1 inhi- bitors, there is an unmet medical need for novel therapies for a sub- group of patients who develop resistance to treatment. Here we employ a wide array of approaches (crystallographic, biochemical, ex vivo cellular, and in vivo) to demonstrate that axitinib selectively targets BCR–ABL1(T315I) through a gatekeeper-mutant-selective mechanism. Although several drug candidates have been reported to target this gate- keeper mutation, including the approved drug ponatinib 20–23, their clin- ical utility has typically been limited by toxicity. Axitinib is a valuable addition to the BCR–ABL1 armamentarium because it is (1) selective towards the gatekeeper mutant, (2) binds to the T315I active site in a distinctive manner as compared to other ABL1 drugs, and (3) has a nar- row target profile14,23 (Extended Data Fig. 4), which could translate into fewer and less severe side effects. This narrow-profile gatekeeper- selective mechanism of action highlights axitinib as a new type of ABL1 kinase inhibitor. This finding could pave the way for a new approach for drug development towards even more potent and selective gate- keeper-mutant inhibitors targeting ABL1 as well as other relevant ki- nases such as EGFR and KIT and their mutation-specific conformations. Since axitinib is already approved for patients with refractory renal cell carcinoma and has manageable side effects24–26, our study provides a solid rationale for formal exploration of the clinical utility of axitinib in drug-resistant BCR–ABL1(T315I)-driven leukaemia in a fast-track mode, probably in combination with a conventional ABL1 inhibitor.

Axitinib may represent a unique opportunity for combinations with other ABL1 drugs because it does not have a similar adverse event pro- file or overlapping toxicities27,28. Furthermore, most current ABL1 drugs cause inactivation of the enzyme that metabolizes them (cytochrome P450, family 3, subfamily A, polypeptide 4) that would complicate the prediction of drug exposures in combinations. Axitinib and bosutinib do not have this issue and their combination may enable broad-spectrum therapy. It is also plausible that the adverse effects of axitinib (mainly VEGFR-driven) could be minimized by intermittent exposure, which may be sufficient for treating BCR–ABL1(T315I)-driven leukaemia as indicated with other ABL1 inhibitors29,30.
In summary, our results on axitinib in T315I mutant CML provide a powerful example of how unbiased drug sensitivity testing of patient- derived cancer cells can lead to the discovery of an unexpected drug– target interaction with mechanistic, structural and clinical implications. Our study highlights the value of drug repositioning, that is, searching for novel indications for existing, emerging and abandoned drugs, such as in the NIH program ‘Discovering New Therapeutic Uses for Existing Molecules’.

METHODS

Study patients and material. Primary patient material was obtained after written informed consent approved by the Helsinki University Central Hospital Institu- tional Review Board (No. 239/13/00/2010, 303/13/03/01/2011). The consent also included the possibility to utilize DSRT data to guide therapies with approved agents in an off-label manner in accordance with Finnish legislation. Mononuclear cells from bone marrow aspirates of CML and Ph1 ALL patients were isolated using Ficoll centrifugation (Ficoll-Paque PREMIUM; GE Healthcare) and maintained in Mononuclear Cell Medium (MCM; PromoCell) supplemented with 0.5 mg ml21 gentamicin and 2.5 mg ml21 amphotericin B. Patient characteristics are summar- ized in Extended Data Table 2.
Inhibitors and DSRT. The oncology compound collection used in this study consisted of 125 FDA/EMA approved anti-cancer drugs, along with 127 investiga- tional and preclinical compounds covering a wide spectrum of molecular targets (Supplementary Table 1). All of the compounds were purchased from commercial chemical vendors and dissolved in either 100% dimethyl sulfoxide (DMSO) or water. The DSRT was performed as previously described12.
Specifically, each compound was tested in five different concentrations covering a 10,000-fold concentration range and preprinted on 384 microtitre tissue culture treated plates (Corning) with an acoustic liquid handling device (Echo 550, Labcyte Inc.). Five microlitres of medium was added to each well for compound dissolution and the plates were gently shaken for 30 min. A single-cell suspension of freshly isolated mononuclear cells (20 ml per well; 10,000 cells per well) was then transferred to every well using Multi- Drop Combi peristaltic dispenser (Thermo Scientific). The 384-well plates were then incubated for 72 h at 37 uC and 5% CO2 and following the incubation period cell viability was measured using CellTiter-Glo reagent (Promega) according to the manufacturer’s instructions with Molecular Devices Paradigm plate reader. Cell viability luminescence data was normalized to DMSO-only wells (negative con- trol) and 100 mM benzethonium-chloride-containing wells (positive control). Dose- response curves were generated in Dotmatics Browser/Studies software (Dotmatics Ltd) on the basis of four-parameter logistics fit function (minimum and maximum inhibition, slope and half maximal inhibitory concentration (IC50)). The DSRT data were evaluated with a custom-developed drug sensitivity score (DSS)12,13. Selective drug responses in the CML and Ph1 ALL patient cells were evaluated in compar- ison with the average control cell drug sensitivity profile, sDSS.

Kinase assays. Wild-type ABL1 and ABL1(T315I) enzyme inhibition was measured using a microfluidic mobility shift assay. The reactions were conducted in 50 ml vol- umes in 96-well plates and contained GST-tagged human-recombinant ABL1(T315I) kinase intracellular domain (1 nM), 3 mM phosphoacceptor peptide, 59 FAM- EAIYAAPFAKKK-OH (CPC Scientific, also known as ProfilerPro Peptide 2, Cal- iper Life Sciences), test compound (11-dose threefold serial dilutions, 2% DMSO final) or DMSO only, 1 mM dithiothreitol (DTT), 0.002% Tween-20 and 5 mM MgCl2 in 25 mM HEPES, pH 7.1. The reactions were initiated by addition of ATP (5 mM final concentration, which is approximately equal to the Michaelis constant (Km)) following a 20 min pre-incubation, incubated for 1.5 h at room temperature and stopped by the addition of 0.1 M EDTA, pH 8. Extent of reaction (,15–20% conversion with no inhibitor) was determined after electrophoretic separation of the fluorescently labelled peptide substrate and phosphorylated product on a LabChip EZ Reader II (Caliper Life Sciences). The Ki values were calculated by fitting the percentage conversion to the Morrison31 equation for tight-binding competitive inhibition using a nonlinear regression method (GraphPad Prism), best fit enzyme concentration value and an experimentally measured ATP Km (,4 mM).

Production of recombinant ABL1. The nucleotide sequence encoding residues 229–515 of human ABL1a (NM_005157.4) was obtained from GenScript sub-cloned into an insect cell expression transfer vector that appended the N-terminus with the tobacco etch virus (TEV)-cleavable polyhistidine purification tag sequence MASHHHHHHDYDGATTENLYFQ/GS, where TEV cleaves at the solidus, leav- ing the recombinant protein with a Gly–Ser extension at the N-terminus. Site- directed mutagenesis was used to generate the T315I mutation within this construct. TEV protease was produced in-house under license from NCI32. Recombinant baculoviruses were prepared by using the ‘Bac-to-Bac’ method (Invitrogen) and used to infect 10 l of Sf21 insect cells at 27 uC and at a multiplicity of infection (MOI) of 1. The ABL viruses were co-infected with a baculovirus expressing YopH tyrosine phosphatase at a MOI of 0.01, in order to generate the non-phosphorylated protein species. Infected cells were harvested 72 h post-infection and the PBS- washed cell pellets were stored at 280 uC before purification.

Frozen insect cell pellets containing recombinant ABL1(T315I) mutant kinase domain protein were resuspended in lysis buffer (50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 10 mM MgCl2, 5 mM ADP-NaOH pH 7.5, 0.25 mM TCEP, 2 mM leupeptin (Sigma Chemical), and one ‘EDTA-free’ protease inhibitor tablet (Roche) per 75 ml buffer), and the mixture stirred at 4 uC for 1 h followed by centrifugation at 5,000g for 1 h. The supernatant fraction was incubated with 5 ml of ProBond resin (Invitrogen) for approximately 3 h with mixing at 4 uC. Subsequently, the resin containing bound ABL1 was batch washed with 4 3 50 ml of wash buffer (50 mM Tris-HCl, pH 8.0, 400 mM NaCl, 20 mM imidazole-HCl, pH 8.0, 10 mM MgCl2, 2.5 mM ADP-NaOH pH 7.5, 0.25 mM TCEP, 1 mM leupeptin) and the resin trans- ferred to a disposable Econo column (Bio-Rad). The resin was washed further with 20 ml of wash buffer and the bound protein step-eluted by using four column volumes of elution buffer (50 mM Tris-HCl, pH 8.0, 400 mM NaCl, 250 mM imidazole- HCl, pH 8.0, 10 mM MgCl2, 2.5 mM ADP-NaOH pH 7.5, 0.25 mM TCEP, 1 mM leupeptin). The eluted protein was treated with TEV protease during overnight dialysis against wash buffer. The dialysed material was passed through a fresh 10 ml column of ProBond resin previously equilibrated with the post-dialysis buffer and the flow-through fraction containing the detagged ABL1(T315I) was collected. The protein solution was dialysed against delivery buffer (25 mM HEPES-NaOH, pH 7.2, 250 mM NaCl, 5 mM MgCl2, 2.5 mM ADP-NaOH, pH 7.5, 20% (v/v) glycerol, 0.25 mM TCEP). The wild-type ABL1 kinase domain was purified in a similar man- ner with the exception that MgCl2 and ADP-NaOH pH 7.0 were omitted from all buffers. The purified proteins at 2–3 mg ml21 were flash frozen in liquid N2 and stored at 280 uC. Protein measurements were determined by using the Coomassie Plus Protein Reagent (Pierce).

Enzyme preparation. Frozen wild-type ABL1 and ABL1(T315I) proteins were thawed and diluted to 10 mM (,0.3 mg ml21) with cold 25 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM DTT. An equal volume of cold buffer containing 20 mM axitinib (freshly diluted from a 20 mM stock in DMSO) was mixed with the pro- tein and placed on ice for 1 h. The protein mixture was concentratedto ,20 mg ml21 by using a Millipore centrifugal concentrator and the proteins flash frozen in liquid N2 before crystallization trials.

Crystallization. All crystals were obtained by sitting drop vapour diffusion in SBS format MRC2 crystallization plates using a Mosquito liquid handler. Crystals of ABL(T315I) were obtained at 13 uC by mixing 196 nl protein–axitinib complex so- lution (17 mg ml21) with 211 nl of reservoir solution (15.0% (w/v) PEG 3350, 10 mM MgCl2, 5 mM NiCl2, 5.0% (v/v) glycerol, and 100 mM HEPES, pH 7). Wild-type ABL1 crystals grew from drops containing 150 nl protein (18 mg ml21) mixed with 190 nl of reservoir solution (0.1 M ammonium chloride, 20.0% (w/v) PEG 3350 and 5.0% (v/v) ethylene glycol). Prior to crystal harvest, the crystallization drop was covered with 4 ml of reservoir solution containing 20% glycerol (T315I) or 20% ethylene glycol (wild type) as cryo-protectant. Crystals were then harvested directly from the crystallization drop followed by immediate flash-freezing in liquid N2. Data collection and structure determination. X-ray data sets were collected at beamline 17-ID at the Advanced Photon Source synchrotron (Argonne National
Laboratories) using a wavelength of 1 A˚ , and at a temperature of 100 K with Rmerge 5 0.055 for data in the range of 65.58–2.40 A˚ for ABL1(T315I), and Rmerge 5 0.057 for data in the range of 111.81–2.20 A˚ for wild-type ABL1 (Ex- tended Data Table 1). Structure solution for ABL1(T315I) proceeded by molecular replacement in PHASER33 using the ABL1 structure (PDB ID: 3IK3) as the starting model followed by iterative rounds of model building and refinement using Coot34 and REFMAC535. Final refinement and structure validation were performed in the PHENIX suite36. Structure solution, model building and refinement for wild-type ABL1 proceeded in an identical manner but using the refined ABL1(T315) atomic coordinates as a starting model for molecular replacement. Crystallographic R factors and stereochemistry statistics indicate high quality models for each refinement (Ex- tended Data Table 1). For ABL1(T315I) and wild-type ABL1 structures respectively Ramachandran plots indicate 97.9% and 97.1% of residues are in the favoured, and 2.1% and 2.9% in the allowed, regions. Atomic coordinates and structure factors for the reported crystal structures have been deposited in the Protein Data Bank under accession numbers 4TWP and 4WA9 for ABL1(T315I) and wild-type ABL1 respectively.

BCR–ABL1 autophosphorylation ELISA. Murine pro-B Ba/F3 cell lines expres- sing human wild-type BCR–ABL1 and mutant BCR–ABL1(T315I) were obtained from Oregon Health and Science University and grown in RPMI-1640 medium sup- plemented with 10% FBS and 1% penicillin–streptomycin. For the ABL1 phospho- Tyr ELISA, cells were pipetted into a 50 ml tube, centrifuged at 180g, and the cell pellet was re-suspended in assay medium (RPMI-1640 with 0.1% FBS, 0.05% BSA w/v, and 1% penicillin–streptomycin). The cells were counted using an Innovatis Cedex Cell Counter, seeded into a 96-well flat-bottom plate in assay medium at 40,000 cells per well and incubated for 2 h at 37 uC, 5% CO2 95% air. Test com- pounds were dissolved in 100% DMSO to make 10 mM stocks. The compounds were then diluted in 100% DMSO in a polypropylene 96-well plate using threefold serial dilution. Control wells contained 100% DMSO without test compound (unin- hibited controls). The DMSO drug dilution plate was diluted 40-fold into assay medium to yield a 53 drug source plate for the assay. Twenty-five microlitres was transferred from the 53 source plate to the cell assay plate and the assay plate was incubated with test compounds for an additional 2 h at 37 uC, 5% CO2 95% air. Following this incubation, the cells were centrifuged at 405g for 5 min and 80 ml of the supernatant was removed. Cells were lysed by adding 100 ml per well of freshly prepared Cell Signaling Technology lysis buffer (#9803) supplemented with 1% SDS, protease inhibitors (Sigma P8340), and phosphatase inhibitors (Sigma P0044 and Sigma P5726). The cell assay plate with lysis buffer was shaken for 10 min at 4 uC and then 100 ml of cell lysate from each well was transferred to a goat anti- rabbit 96-well ELISA plate (Pierce #15135), which was previously incubated with rabbit anti-ABL1 antibody (Cell Signaling Technology #2862) diluted 1:200 in blocking buffer (Pierce StartingBlock). The cell lysate was incubated with the anti- c-ABL1-coated ELISA plate for 1 h at room temperature and then washed four times with Cell Signaling Technology ELISA Wash Buffer (from kit #7903). The final wash was removed by inverting the plate. One hundred microlitres of mouse monoclonal (IgG2b) anti-phospho-Tyr antibody (Santa Cruz Biotechnology SC508 HRP) diluted 1:5,000 was added to each well. The ELISA plate was then incubated for 45 min at room temperature with 100 ml per well. The plate was washed four times as described above, the final wash removed, and 100 ml of TMB substrate (Santa Cruz Biotechnology SC286967) was added to each well. Absorbance was measured at 655 nm during colour development or the reaction stopped by adding 50–100 ml per well of 0.16 M sulfuric acid stop solution and read at 450 nm.

BCR–ABL Ba/F3 proliferation assay. For the proliferation assay, Ba/F3 cells were pipetted into a 50 ml tube, centrifuged at 180g, and the cell pellet was re-suspended in RPMI-1640 with 1% FBS, and 1% penicillin–streptomycin. The cells were counted using an Innovatis Cedex Cell Counter and seeded into a 96-well flat-bottom plate at 1,500 cells per well. Compounds were serially diluted in 100% DMSO as de- scribed above and then diluted 40-fold into RPMI-1640 with 1% FBS and 1% penicillin–streptomycin to yield a 53 source plate. Twenty-five microlitres was transferred from the 53 source plate to the cell assay plate and the cells incubated with test compounds for 4 days at 37 uC, 5% CO2 95% air. On day 4 post drug addi- tion, the cell assay plate was centrifuged at 180g for 2 min, 80 ml of supernatant was removed from each well, and 100 ml of fresh medium was added to each well. Fif- teen microlitres of 1 mg ml21 Resazurin (Sigma R7017) was then added to each well and the cell assay plate was incubated for 6 h at 37 uC, 5% CO2 95% air. The fluo- rescence signal was read using 530 nm excitation and 595 nm emission wavelengths. Engineered Ba/F3 cell proliferation method used in Extended Data Fig. 1. Cells were seeded at a concentration of 10,000 cells per well (10% FBS) in 96-well round- bottom cell culture plates with complete medium and in the presence of increasing concentrations of axitinib (range 0–10 mM). Cell proliferation was measured at 72 h using the tritiated thymidine incorporation assay as described previously37. The labelling time was for 8 h (started 64 h after seeding). Each test was performed in quadruplicate and replicated at least twice. Calculation of IC50 values was per- formed using GraphPad Prism software.

CRKL phosphorylation in CML patient cells. Mononuclear cells (5 3 106 per condition) of a CML patient sample (FHRB.1408) harbouring the T315I mutation (confirmed by reverse-transcription PCR and sequencing) were cultured overnight in either complete medium plus 0.1% DMSO or increasing concentrations of axi- tinib (1–1,000 nM; tenfold dilutions). Following incubation the cells were centrifuged, washed with cold PBS and lysed in 4% SDS, 0.1 M DTT and 0.1 M Tris. Lysate proteins were separated by SDS PAGE and transferred to an Immobilon-FL poly- vinylidene difluoride (PVDF) membrane (Millipore). The membrane was blocked with 5% bovine serum albumin for 1 h, and incubated with phospho-CRKL (Y207; Cell Signaling Technologies (3181S); 1:1,000 dilution) and a-tubulin (Sigma Aldrich; T9026 mouse mAb; 1:1,000 dilution) antibodies in 5% BSA Tris-buffered saline and 0.05% Tween 20 (TBS-T) overnight at 4 uC. The membranes were then incubated with secondary infrared-labelled antibodies anti-mouse IRDye 680 and anti-rabbit IRDye 800CW (Odyssey; LI-COR Biosciences; 1:15,000 dilution) for 1 h at room temperature. The protein bands were visualized with the Odyssey imaging system (LI-COR Biosciences).

BCR–ABL1(T315I) transcript quantification. The proportion of T315I-mutation- positive transcripts was assessed with RT–qPCR using a mutation-specific forward primer, a reverse primer and a fluorescent TaqMan probe. The patients’ bone mar- row complementary DNA sample, that was found T315I-mutation-positive by Sanger sequencing, served as quantification standard for the follow-up samples. The cDNA from the T315I-positive sample was diluted into a negative control cDNA in a log-linear fashion to construct a dilution series and a corresponding standard curve. The standard curve was used for quantifying the T315I level in the samples taken at treatment day 0 and at day 14 (Fig. 4e). The results were normalized using GUS reference gene to compensate the differences inthe RNA quality and the cDNA synthesis. Based on repeated measurements of the quantification standard in four independent runs, including two replicate analyses in each run, the between-run coefficient of variation for the measurement of the T315I:GUS ratio was 19.2%. The treatment time point samples were each measured by using three replicate analyses and an unpaired two-tailed t-test with Welch’s correction indicated significant dif- ference between the samples (P , 0.05).